Linux Audio

Check our new training course

Loading...
v6.8
  1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 *
  4 *  Copyright (C) 1995  Linus Torvalds
  5 *
  6 *  Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
  7 */
  8
  9#include <linux/signal.h>
 10#include <linux/sched.h>
 11#include <linux/kernel.h>
 12#include <linux/errno.h>
 13#include <linux/string.h>
 14#include <linux/types.h>
 15#include <linux/ptrace.h>
 16#include <linux/mman.h>
 17#include <linux/mm.h>
 18#include <linux/hugetlb.h>
 19#include <linux/swap.h>
 20#include <linux/smp.h>
 21#include <linux/init.h>
 22#include <linux/highmem.h>
 23#include <linux/pagemap.h>
 24#include <linux/pci.h>
 25#include <linux/pfn.h>
 26#include <linux/poison.h>
 27#include <linux/memblock.h>
 28#include <linux/proc_fs.h>
 29#include <linux/memory_hotplug.h>
 30#include <linux/initrd.h>
 31#include <linux/cpumask.h>
 32#include <linux/gfp.h>
 33
 34#include <asm/asm.h>
 35#include <asm/bios_ebda.h>
 36#include <asm/processor.h>
 37#include <linux/uaccess.h>
 
 38#include <asm/dma.h>
 39#include <asm/fixmap.h>
 40#include <asm/e820/api.h>
 41#include <asm/apic.h>
 42#include <asm/bugs.h>
 43#include <asm/tlb.h>
 44#include <asm/tlbflush.h>
 45#include <asm/olpc_ofw.h>
 46#include <asm/pgalloc.h>
 47#include <asm/sections.h>
 
 48#include <asm/setup.h>
 49#include <asm/set_memory.h>
 50#include <asm/page_types.h>
 51#include <asm/cpu_entry_area.h>
 52#include <asm/init.h>
 53#include <asm/pgtable_areas.h>
 54#include <asm/numa.h>
 55
 56#include "mm_internal.h"
 57
 58unsigned long highstart_pfn, highend_pfn;
 59
 60bool __read_mostly __vmalloc_start_set = false;
 61
 62/*
 63 * Creates a middle page table and puts a pointer to it in the
 64 * given global directory entry. This only returns the gd entry
 65 * in non-PAE compilation mode, since the middle layer is folded.
 66 */
 67static pmd_t * __init one_md_table_init(pgd_t *pgd)
 68{
 69	p4d_t *p4d;
 70	pud_t *pud;
 71	pmd_t *pmd_table;
 72
 73#ifdef CONFIG_X86_PAE
 74	if (!(pgd_val(*pgd) & _PAGE_PRESENT)) {
 75		pmd_table = (pmd_t *)alloc_low_page();
 
 76		set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
 77		p4d = p4d_offset(pgd, 0);
 78		pud = pud_offset(p4d, 0);
 79		BUG_ON(pmd_table != pmd_offset(pud, 0));
 80
 81		return pmd_table;
 82	}
 83#endif
 84	p4d = p4d_offset(pgd, 0);
 85	pud = pud_offset(p4d, 0);
 86	pmd_table = pmd_offset(pud, 0);
 87
 88	return pmd_table;
 89}
 90
 91/*
 92 * Create a page table and place a pointer to it in a middle page
 93 * directory entry:
 94 */
 95static pte_t * __init one_page_table_init(pmd_t *pmd)
 96{
 97	if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
 98		pte_t *page_table = (pte_t *)alloc_low_page();
 99
 
100		set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
101		BUG_ON(page_table != pte_offset_kernel(pmd, 0));
102	}
103
104	return pte_offset_kernel(pmd, 0);
105}
106
107pmd_t * __init populate_extra_pmd(unsigned long vaddr)
108{
109	int pgd_idx = pgd_index(vaddr);
110	int pmd_idx = pmd_index(vaddr);
111
112	return one_md_table_init(swapper_pg_dir + pgd_idx) + pmd_idx;
113}
114
115pte_t * __init populate_extra_pte(unsigned long vaddr)
116{
117	int pte_idx = pte_index(vaddr);
118	pmd_t *pmd;
119
120	pmd = populate_extra_pmd(vaddr);
121	return one_page_table_init(pmd) + pte_idx;
122}
123
124static unsigned long __init
125page_table_range_init_count(unsigned long start, unsigned long end)
126{
127	unsigned long count = 0;
128#ifdef CONFIG_HIGHMEM
129	int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
130	int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
131	int pgd_idx, pmd_idx;
132	unsigned long vaddr;
133
134	if (pmd_idx_kmap_begin == pmd_idx_kmap_end)
135		return 0;
136
137	vaddr = start;
138	pgd_idx = pgd_index(vaddr);
139	pmd_idx = pmd_index(vaddr);
140
141	for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd_idx++) {
142		for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
143							pmd_idx++) {
144			if ((vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin &&
145			    (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end)
146				count++;
147			vaddr += PMD_SIZE;
148		}
149		pmd_idx = 0;
150	}
151#endif
152	return count;
153}
154
155static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd,
156					   unsigned long vaddr, pte_t *lastpte,
157					   void **adr)
158{
159#ifdef CONFIG_HIGHMEM
160	/*
161	 * Something (early fixmap) may already have put a pte
162	 * page here, which causes the page table allocation
163	 * to become nonlinear. Attempt to fix it, and if it
164	 * is still nonlinear then we have to bug.
165	 */
166	int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
167	int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
168
169	if (pmd_idx_kmap_begin != pmd_idx_kmap_end
170	    && (vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin
171	    && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end) {
172		pte_t *newpte;
173		int i;
174
175		BUG_ON(after_bootmem);
176		newpte = *adr;
177		for (i = 0; i < PTRS_PER_PTE; i++)
178			set_pte(newpte + i, pte[i]);
179		*adr = (void *)(((unsigned long)(*adr)) + PAGE_SIZE);
180
 
181		set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE));
182		BUG_ON(newpte != pte_offset_kernel(pmd, 0));
183		__flush_tlb_all();
184
 
185		pte = newpte;
186	}
187	BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1)
188	       && vaddr > fix_to_virt(FIX_KMAP_END)
189	       && lastpte && lastpte + PTRS_PER_PTE != pte);
190#endif
191	return pte;
192}
193
194/*
195 * This function initializes a certain range of kernel virtual memory
196 * with new bootmem page tables, everywhere page tables are missing in
197 * the given range.
198 *
199 * NOTE: The pagetables are allocated contiguous on the physical space
200 * so we can cache the place of the first one and move around without
201 * checking the pgd every time.
202 */
203static void __init
204page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
205{
206	int pgd_idx, pmd_idx;
207	unsigned long vaddr;
208	pgd_t *pgd;
209	pmd_t *pmd;
210	pte_t *pte = NULL;
211	unsigned long count = page_table_range_init_count(start, end);
212	void *adr = NULL;
213
214	if (count)
215		adr = alloc_low_pages(count);
216
217	vaddr = start;
218	pgd_idx = pgd_index(vaddr);
219	pmd_idx = pmd_index(vaddr);
220	pgd = pgd_base + pgd_idx;
221
222	for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
223		pmd = one_md_table_init(pgd);
224		pmd = pmd + pmd_index(vaddr);
225		for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
226							pmd++, pmd_idx++) {
227			pte = page_table_kmap_check(one_page_table_init(pmd),
228						    pmd, vaddr, pte, &adr);
229
230			vaddr += PMD_SIZE;
231		}
232		pmd_idx = 0;
233	}
234}
235
236static inline int is_x86_32_kernel_text(unsigned long addr)
237{
238	if (addr >= (unsigned long)_text && addr <= (unsigned long)__init_end)
239		return 1;
240	return 0;
241}
242
243/*
244 * This maps the physical memory to kernel virtual address space, a total
245 * of max_low_pfn pages, by creating page tables starting from address
246 * PAGE_OFFSET:
247 */
248unsigned long __init
249kernel_physical_mapping_init(unsigned long start,
250			     unsigned long end,
251			     unsigned long page_size_mask,
252			     pgprot_t prot)
253{
254	int use_pse = page_size_mask == (1<<PG_LEVEL_2M);
255	unsigned long last_map_addr = end;
256	unsigned long start_pfn, end_pfn;
257	pgd_t *pgd_base = swapper_pg_dir;
258	int pgd_idx, pmd_idx, pte_ofs;
259	unsigned long pfn;
260	pgd_t *pgd;
261	pmd_t *pmd;
262	pte_t *pte;
263	unsigned pages_2m, pages_4k;
264	int mapping_iter;
265
266	start_pfn = start >> PAGE_SHIFT;
267	end_pfn = end >> PAGE_SHIFT;
268
269	/*
270	 * First iteration will setup identity mapping using large/small pages
271	 * based on use_pse, with other attributes same as set by
272	 * the early code in head_32.S
273	 *
274	 * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
275	 * as desired for the kernel identity mapping.
276	 *
277	 * This two pass mechanism conforms to the TLB app note which says:
278	 *
279	 *     "Software should not write to a paging-structure entry in a way
280	 *      that would change, for any linear address, both the page size
281	 *      and either the page frame or attributes."
282	 */
283	mapping_iter = 1;
284
285	if (!boot_cpu_has(X86_FEATURE_PSE))
286		use_pse = 0;
287
288repeat:
289	pages_2m = pages_4k = 0;
290	pfn = start_pfn;
291	pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
292	pgd = pgd_base + pgd_idx;
293	for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
294		pmd = one_md_table_init(pgd);
295
296		if (pfn >= end_pfn)
297			continue;
298#ifdef CONFIG_X86_PAE
299		pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
300		pmd += pmd_idx;
301#else
302		pmd_idx = 0;
303#endif
304		for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn;
305		     pmd++, pmd_idx++) {
306			unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
307
308			/*
309			 * Map with big pages if possible, otherwise
310			 * create normal page tables:
311			 */
312			if (use_pse) {
313				unsigned int addr2;
314				pgprot_t prot = PAGE_KERNEL_LARGE;
315				/*
316				 * first pass will use the same initial
317				 * identity mapping attribute + _PAGE_PSE.
318				 */
319				pgprot_t init_prot =
320					__pgprot(PTE_IDENT_ATTR |
321						 _PAGE_PSE);
322
323				pfn &= PMD_MASK >> PAGE_SHIFT;
324				addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
325					PAGE_OFFSET + PAGE_SIZE-1;
326
327				if (is_x86_32_kernel_text(addr) ||
328				    is_x86_32_kernel_text(addr2))
329					prot = PAGE_KERNEL_LARGE_EXEC;
330
331				pages_2m++;
332				if (mapping_iter == 1)
333					set_pmd(pmd, pfn_pmd(pfn, init_prot));
334				else
335					set_pmd(pmd, pfn_pmd(pfn, prot));
336
337				pfn += PTRS_PER_PTE;
338				continue;
339			}
340			pte = one_page_table_init(pmd);
341
342			pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
343			pte += pte_ofs;
344			for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn;
345			     pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
346				pgprot_t prot = PAGE_KERNEL;
347				/*
348				 * first pass will use the same initial
349				 * identity mapping attribute.
350				 */
351				pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR);
352
353				if (is_x86_32_kernel_text(addr))
354					prot = PAGE_KERNEL_EXEC;
355
356				pages_4k++;
357				if (mapping_iter == 1) {
358					set_pte(pte, pfn_pte(pfn, init_prot));
359					last_map_addr = (pfn << PAGE_SHIFT) + PAGE_SIZE;
360				} else
361					set_pte(pte, pfn_pte(pfn, prot));
362			}
363		}
364	}
365	if (mapping_iter == 1) {
366		/*
367		 * update direct mapping page count only in the first
368		 * iteration.
369		 */
370		update_page_count(PG_LEVEL_2M, pages_2m);
371		update_page_count(PG_LEVEL_4K, pages_4k);
372
373		/*
374		 * local global flush tlb, which will flush the previous
375		 * mappings present in both small and large page TLB's.
376		 */
377		__flush_tlb_all();
378
379		/*
380		 * Second iteration will set the actual desired PTE attributes.
381		 */
382		mapping_iter = 2;
383		goto repeat;
384	}
385	return last_map_addr;
386}
387
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
388#ifdef CONFIG_HIGHMEM
389static void __init permanent_kmaps_init(pgd_t *pgd_base)
390{
391	unsigned long vaddr = PKMAP_BASE;
 
 
 
 
 
392
 
393	page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
394
395	pkmap_page_table = virt_to_kpte(vaddr);
 
 
 
 
 
396}
397
398void __init add_highpages_with_active_regions(int nid,
399			 unsigned long start_pfn, unsigned long end_pfn)
400{
401	phys_addr_t start, end;
402	u64 i;
403
404	for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &start, &end, NULL) {
405		unsigned long pfn = clamp_t(unsigned long, PFN_UP(start),
406					    start_pfn, end_pfn);
407		unsigned long e_pfn = clamp_t(unsigned long, PFN_DOWN(end),
408					      start_pfn, end_pfn);
409		for ( ; pfn < e_pfn; pfn++)
410			if (pfn_valid(pfn))
411				free_highmem_page(pfn_to_page(pfn));
412	}
413}
414#else
415static inline void permanent_kmaps_init(pgd_t *pgd_base)
416{
417}
418#endif /* CONFIG_HIGHMEM */
419
420void __init sync_initial_page_table(void)
421{
422	clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY,
423			swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
424			KERNEL_PGD_PTRS);
425
426	/*
427	 * sync back low identity map too.  It is used for example
428	 * in the 32-bit EFI stub.
429	 */
430	clone_pgd_range(initial_page_table,
431			swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
432			min(KERNEL_PGD_PTRS, KERNEL_PGD_BOUNDARY));
433}
434
435void __init native_pagetable_init(void)
436{
437	unsigned long pfn, va;
438	pgd_t *pgd, *base = swapper_pg_dir;
439	p4d_t *p4d;
440	pud_t *pud;
441	pmd_t *pmd;
442	pte_t *pte;
443
444	/*
445	 * Remove any mappings which extend past the end of physical
446	 * memory from the boot time page table.
447	 * In virtual address space, we should have at least two pages
448	 * from VMALLOC_END to pkmap or fixmap according to VMALLOC_END
449	 * definition. And max_low_pfn is set to VMALLOC_END physical
450	 * address. If initial memory mapping is doing right job, we
451	 * should have pte used near max_low_pfn or one pmd is not present.
452	 */
453	for (pfn = max_low_pfn; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
454		va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
455		pgd = base + pgd_index(va);
456		if (!pgd_present(*pgd))
457			break;
458
459		p4d = p4d_offset(pgd, va);
460		pud = pud_offset(p4d, va);
461		pmd = pmd_offset(pud, va);
462		if (!pmd_present(*pmd))
463			break;
464
465		/* should not be large page here */
466		if (pmd_large(*pmd)) {
467			pr_warn("try to clear pte for ram above max_low_pfn: pfn: %lx pmd: %p pmd phys: %lx, but pmd is big page and is not using pte !\n",
468				pfn, pmd, __pa(pmd));
469			BUG_ON(1);
470		}
471
472		pte = pte_offset_kernel(pmd, va);
473		if (!pte_present(*pte))
474			break;
475
476		printk(KERN_DEBUG "clearing pte for ram above max_low_pfn: pfn: %lx pmd: %p pmd phys: %lx pte: %p pte phys: %lx\n",
477				pfn, pmd, __pa(pmd), pte, __pa(pte));
478		pte_clear(NULL, va, pte);
479	}
 
480	paging_init();
481}
482
483/*
484 * Build a proper pagetable for the kernel mappings.  Up until this
485 * point, we've been running on some set of pagetables constructed by
486 * the boot process.
487 *
488 * This will be a pagetable constructed in arch/x86/kernel/head_32.S.
489 * The root of the pagetable will be swapper_pg_dir.
 
 
 
 
 
 
 
490 *
491 * In general, pagetable_init() assumes that the pagetable may already
492 * be partially populated, and so it avoids stomping on any existing
493 * mappings.
494 */
495void __init early_ioremap_page_table_range_init(void)
496{
497	pgd_t *pgd_base = swapper_pg_dir;
498	unsigned long vaddr, end;
499
500	/*
501	 * Fixed mappings, only the page table structure has to be
502	 * created - mappings will be set by set_fixmap():
503	 */
504	vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
505	end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
506	page_table_range_init(vaddr, end, pgd_base);
507	early_ioremap_reset();
508}
509
510static void __init pagetable_init(void)
511{
512	pgd_t *pgd_base = swapper_pg_dir;
513
514	permanent_kmaps_init(pgd_base);
515}
516
517#define DEFAULT_PTE_MASK ~(_PAGE_NX | _PAGE_GLOBAL)
518/* Bits supported by the hardware: */
519pteval_t __supported_pte_mask __read_mostly = DEFAULT_PTE_MASK;
520/* Bits allowed in normal kernel mappings: */
521pteval_t __default_kernel_pte_mask __read_mostly = DEFAULT_PTE_MASK;
522EXPORT_SYMBOL_GPL(__supported_pte_mask);
523/* Used in PAGE_KERNEL_* macros which are reasonably used out-of-tree: */
524EXPORT_SYMBOL(__default_kernel_pte_mask);
525
526/* user-defined highmem size */
527static unsigned int highmem_pages = -1;
528
529/*
530 * highmem=size forces highmem to be exactly 'size' bytes.
531 * This works even on boxes that have no highmem otherwise.
532 * This also works to reduce highmem size on bigger boxes.
533 */
534static int __init parse_highmem(char *arg)
535{
536	if (!arg)
537		return -EINVAL;
538
539	highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
540	return 0;
541}
542early_param("highmem", parse_highmem);
543
544#define MSG_HIGHMEM_TOO_BIG \
545	"highmem size (%luMB) is bigger than pages available (%luMB)!\n"
546
547#define MSG_LOWMEM_TOO_SMALL \
548	"highmem size (%luMB) results in <64MB lowmem, ignoring it!\n"
549/*
550 * All of RAM fits into lowmem - but if user wants highmem
551 * artificially via the highmem=x boot parameter then create
552 * it:
553 */
554static void __init lowmem_pfn_init(void)
555{
556	/* max_low_pfn is 0, we already have early_res support */
557	max_low_pfn = max_pfn;
558
559	if (highmem_pages == -1)
560		highmem_pages = 0;
561#ifdef CONFIG_HIGHMEM
562	if (highmem_pages >= max_pfn) {
563		printk(KERN_ERR MSG_HIGHMEM_TOO_BIG,
564			pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
565		highmem_pages = 0;
566	}
567	if (highmem_pages) {
568		if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) {
569			printk(KERN_ERR MSG_LOWMEM_TOO_SMALL,
570				pages_to_mb(highmem_pages));
571			highmem_pages = 0;
572		}
573		max_low_pfn -= highmem_pages;
574	}
575#else
576	if (highmem_pages)
577		printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
578#endif
579}
580
581#define MSG_HIGHMEM_TOO_SMALL \
582	"only %luMB highmem pages available, ignoring highmem size of %luMB!\n"
583
584#define MSG_HIGHMEM_TRIMMED \
585	"Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n"
586/*
587 * We have more RAM than fits into lowmem - we try to put it into
588 * highmem, also taking the highmem=x boot parameter into account:
589 */
590static void __init highmem_pfn_init(void)
591{
592	max_low_pfn = MAXMEM_PFN;
593
594	if (highmem_pages == -1)
595		highmem_pages = max_pfn - MAXMEM_PFN;
596
597	if (highmem_pages + MAXMEM_PFN < max_pfn)
598		max_pfn = MAXMEM_PFN + highmem_pages;
599
600	if (highmem_pages + MAXMEM_PFN > max_pfn) {
601		printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL,
602			pages_to_mb(max_pfn - MAXMEM_PFN),
603			pages_to_mb(highmem_pages));
604		highmem_pages = 0;
605	}
606#ifndef CONFIG_HIGHMEM
607	/* Maximum memory usable is what is directly addressable */
608	printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20);
609	if (max_pfn > MAX_NONPAE_PFN)
610		printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
611	else
612		printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
613	max_pfn = MAXMEM_PFN;
614#else /* !CONFIG_HIGHMEM */
615#ifndef CONFIG_HIGHMEM64G
616	if (max_pfn > MAX_NONPAE_PFN) {
617		max_pfn = MAX_NONPAE_PFN;
618		printk(KERN_WARNING MSG_HIGHMEM_TRIMMED);
619	}
620#endif /* !CONFIG_HIGHMEM64G */
621#endif /* !CONFIG_HIGHMEM */
622}
623
624/*
625 * Determine low and high memory ranges:
626 */
627void __init find_low_pfn_range(void)
628{
629	/* it could update max_pfn */
630
631	if (max_pfn <= MAXMEM_PFN)
632		lowmem_pfn_init();
633	else
634		highmem_pfn_init();
635}
636
637#ifndef CONFIG_NUMA
638void __init initmem_init(void)
639{
640#ifdef CONFIG_HIGHMEM
641	highstart_pfn = highend_pfn = max_pfn;
642	if (max_pfn > max_low_pfn)
643		highstart_pfn = max_low_pfn;
644	printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
645		pages_to_mb(highend_pfn - highstart_pfn));
646	high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
647#else
648	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
649#endif
650
651	memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
 
652
653#ifdef CONFIG_FLATMEM
654	max_mapnr = IS_ENABLED(CONFIG_HIGHMEM) ? highend_pfn : max_low_pfn;
655#endif
656	__vmalloc_start_set = true;
657
658	printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
659			pages_to_mb(max_low_pfn));
660
661	setup_bootmem_allocator();
662}
663#endif /* !CONFIG_NUMA */
664
665void __init setup_bootmem_allocator(void)
666{
667	printk(KERN_INFO "  mapped low ram: 0 - %08lx\n",
668		 max_pfn_mapped<<PAGE_SHIFT);
669	printk(KERN_INFO "  low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT);
670}
671
672/*
673 * paging_init() sets up the page tables - note that the first 8MB are
674 * already mapped by head.S.
675 *
676 * This routines also unmaps the page at virtual kernel address 0, so
677 * that we can trap those pesky NULL-reference errors in the kernel.
678 */
679void __init paging_init(void)
680{
681	pagetable_init();
682
683	__flush_tlb_all();
684
 
 
685	/*
686	 * NOTE: at this point the bootmem allocator is fully available.
687	 */
688	olpc_dt_build_devicetree();
 
689	sparse_init();
690	zone_sizes_init();
691}
692
693/*
694 * Test if the WP bit works in supervisor mode. It isn't supported on 386's
695 * and also on some strange 486's. All 586+'s are OK. This used to involve
696 * black magic jumps to work around some nasty CPU bugs, but fortunately the
697 * switch to using exceptions got rid of all that.
698 */
699static void __init test_wp_bit(void)
700{
701	char z = 0;
702
703	printk(KERN_INFO "Checking if this processor honours the WP bit even in supervisor mode...");
704
705	__set_fixmap(FIX_WP_TEST, __pa_symbol(empty_zero_page), PAGE_KERNEL_RO);
706
707	if (copy_to_kernel_nofault((char *)fix_to_virt(FIX_WP_TEST), &z, 1)) {
708		clear_fixmap(FIX_WP_TEST);
709		printk(KERN_CONT "Ok.\n");
710		return;
711	}
712
713	printk(KERN_CONT "No.\n");
714	panic("Linux doesn't support CPUs with broken WP.");
715}
716
717void __init mem_init(void)
718{
719	pci_iommu_alloc();
720
721#ifdef CONFIG_FLATMEM
722	BUG_ON(!mem_map);
723#endif
724	/*
725	 * With CONFIG_DEBUG_PAGEALLOC initialization of highmem pages has to
726	 * be done before memblock_free_all(). Memblock use free low memory for
727	 * temporary data (see find_range_array()) and for this purpose can use
728	 * pages that was already passed to the buddy allocator, hence marked as
729	 * not accessible in the page tables when compiled with
730	 * CONFIG_DEBUG_PAGEALLOC. Otherwise order of initialization is not
731	 * important here.
732	 */
733	set_highmem_pages_init();
734
735	/* this will put all low memory onto the freelists */
736	memblock_free_all();
737
738	after_bootmem = 1;
739	x86_init.hyper.init_after_bootmem();
740
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
741	/*
742	 * Check boundaries twice: Some fundamental inconsistencies can
743	 * be detected at build time already.
744	 */
745#define __FIXADDR_TOP (-PAGE_SIZE)
746#ifdef CONFIG_HIGHMEM
747	BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE	> FIXADDR_START);
748	BUILD_BUG_ON(VMALLOC_END			> PKMAP_BASE);
749#endif
750#define high_memory (-128UL << 20)
751	BUILD_BUG_ON(VMALLOC_START			>= VMALLOC_END);
752#undef high_memory
753#undef __FIXADDR_TOP
754
755#ifdef CONFIG_HIGHMEM
756	BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE	> FIXADDR_START);
757	BUG_ON(VMALLOC_END				> PKMAP_BASE);
758#endif
759	BUG_ON(VMALLOC_START				>= VMALLOC_END);
760	BUG_ON((unsigned long)high_memory		> VMALLOC_START);
761
762	test_wp_bit();
763}
764
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
765int kernel_set_to_readonly __read_mostly;
766
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
767static void mark_nxdata_nx(void)
768{
769	/*
770	 * When this called, init has already been executed and released,
771	 * so everything past _etext should be NX.
772	 */
773	unsigned long start = PFN_ALIGN(_etext);
774	/*
775	 * This comes from is_x86_32_kernel_text upper limit. Also HPAGE where used:
776	 */
777	unsigned long size = (((unsigned long)__init_end + HPAGE_SIZE) & HPAGE_MASK) - start;
778
779	if (__supported_pte_mask & _PAGE_NX)
780		printk(KERN_INFO "NX-protecting the kernel data: %luk\n", size >> 10);
781	set_memory_nx(start, size >> PAGE_SHIFT);
782}
783
784void mark_rodata_ro(void)
785{
786	unsigned long start = PFN_ALIGN(_text);
787	unsigned long size = (unsigned long)__end_rodata - start;
788
789	set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
790	pr_info("Write protecting kernel text and read-only data: %luk\n",
791		size >> 10);
792
793	kernel_set_to_readonly = 1;
794
795#ifdef CONFIG_CPA_DEBUG
796	pr_info("Testing CPA: Reverting %lx-%lx\n", start, start + size);
797	set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
798
799	pr_info("Testing CPA: write protecting again\n");
800	set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
801#endif
802	mark_nxdata_nx();
803	if (__supported_pte_mask & _PAGE_NX)
804		debug_checkwx();
805}
v5.4
  1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 *
  4 *  Copyright (C) 1995  Linus Torvalds
  5 *
  6 *  Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
  7 */
  8
  9#include <linux/signal.h>
 10#include <linux/sched.h>
 11#include <linux/kernel.h>
 12#include <linux/errno.h>
 13#include <linux/string.h>
 14#include <linux/types.h>
 15#include <linux/ptrace.h>
 16#include <linux/mman.h>
 17#include <linux/mm.h>
 18#include <linux/hugetlb.h>
 19#include <linux/swap.h>
 20#include <linux/smp.h>
 21#include <linux/init.h>
 22#include <linux/highmem.h>
 23#include <linux/pagemap.h>
 24#include <linux/pci.h>
 25#include <linux/pfn.h>
 26#include <linux/poison.h>
 27#include <linux/memblock.h>
 28#include <linux/proc_fs.h>
 29#include <linux/memory_hotplug.h>
 30#include <linux/initrd.h>
 31#include <linux/cpumask.h>
 32#include <linux/gfp.h>
 33
 34#include <asm/asm.h>
 35#include <asm/bios_ebda.h>
 36#include <asm/processor.h>
 37#include <linux/uaccess.h>
 38#include <asm/pgtable.h>
 39#include <asm/dma.h>
 40#include <asm/fixmap.h>
 41#include <asm/e820/api.h>
 42#include <asm/apic.h>
 43#include <asm/bugs.h>
 44#include <asm/tlb.h>
 45#include <asm/tlbflush.h>
 46#include <asm/olpc_ofw.h>
 47#include <asm/pgalloc.h>
 48#include <asm/sections.h>
 49#include <asm/paravirt.h>
 50#include <asm/setup.h>
 51#include <asm/set_memory.h>
 52#include <asm/page_types.h>
 53#include <asm/cpu_entry_area.h>
 54#include <asm/init.h>
 
 
 55
 56#include "mm_internal.h"
 57
 58unsigned long highstart_pfn, highend_pfn;
 59
 60bool __read_mostly __vmalloc_start_set = false;
 61
 62/*
 63 * Creates a middle page table and puts a pointer to it in the
 64 * given global directory entry. This only returns the gd entry
 65 * in non-PAE compilation mode, since the middle layer is folded.
 66 */
 67static pmd_t * __init one_md_table_init(pgd_t *pgd)
 68{
 69	p4d_t *p4d;
 70	pud_t *pud;
 71	pmd_t *pmd_table;
 72
 73#ifdef CONFIG_X86_PAE
 74	if (!(pgd_val(*pgd) & _PAGE_PRESENT)) {
 75		pmd_table = (pmd_t *)alloc_low_page();
 76		paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT);
 77		set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
 78		p4d = p4d_offset(pgd, 0);
 79		pud = pud_offset(p4d, 0);
 80		BUG_ON(pmd_table != pmd_offset(pud, 0));
 81
 82		return pmd_table;
 83	}
 84#endif
 85	p4d = p4d_offset(pgd, 0);
 86	pud = pud_offset(p4d, 0);
 87	pmd_table = pmd_offset(pud, 0);
 88
 89	return pmd_table;
 90}
 91
 92/*
 93 * Create a page table and place a pointer to it in a middle page
 94 * directory entry:
 95 */
 96static pte_t * __init one_page_table_init(pmd_t *pmd)
 97{
 98	if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
 99		pte_t *page_table = (pte_t *)alloc_low_page();
100
101		paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT);
102		set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
103		BUG_ON(page_table != pte_offset_kernel(pmd, 0));
104	}
105
106	return pte_offset_kernel(pmd, 0);
107}
108
109pmd_t * __init populate_extra_pmd(unsigned long vaddr)
110{
111	int pgd_idx = pgd_index(vaddr);
112	int pmd_idx = pmd_index(vaddr);
113
114	return one_md_table_init(swapper_pg_dir + pgd_idx) + pmd_idx;
115}
116
117pte_t * __init populate_extra_pte(unsigned long vaddr)
118{
119	int pte_idx = pte_index(vaddr);
120	pmd_t *pmd;
121
122	pmd = populate_extra_pmd(vaddr);
123	return one_page_table_init(pmd) + pte_idx;
124}
125
126static unsigned long __init
127page_table_range_init_count(unsigned long start, unsigned long end)
128{
129	unsigned long count = 0;
130#ifdef CONFIG_HIGHMEM
131	int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
132	int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
133	int pgd_idx, pmd_idx;
134	unsigned long vaddr;
135
136	if (pmd_idx_kmap_begin == pmd_idx_kmap_end)
137		return 0;
138
139	vaddr = start;
140	pgd_idx = pgd_index(vaddr);
141	pmd_idx = pmd_index(vaddr);
142
143	for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd_idx++) {
144		for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
145							pmd_idx++) {
146			if ((vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin &&
147			    (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end)
148				count++;
149			vaddr += PMD_SIZE;
150		}
151		pmd_idx = 0;
152	}
153#endif
154	return count;
155}
156
157static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd,
158					   unsigned long vaddr, pte_t *lastpte,
159					   void **adr)
160{
161#ifdef CONFIG_HIGHMEM
162	/*
163	 * Something (early fixmap) may already have put a pte
164	 * page here, which causes the page table allocation
165	 * to become nonlinear. Attempt to fix it, and if it
166	 * is still nonlinear then we have to bug.
167	 */
168	int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
169	int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
170
171	if (pmd_idx_kmap_begin != pmd_idx_kmap_end
172	    && (vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin
173	    && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end) {
174		pte_t *newpte;
175		int i;
176
177		BUG_ON(after_bootmem);
178		newpte = *adr;
179		for (i = 0; i < PTRS_PER_PTE; i++)
180			set_pte(newpte + i, pte[i]);
181		*adr = (void *)(((unsigned long)(*adr)) + PAGE_SIZE);
182
183		paravirt_alloc_pte(&init_mm, __pa(newpte) >> PAGE_SHIFT);
184		set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE));
185		BUG_ON(newpte != pte_offset_kernel(pmd, 0));
186		__flush_tlb_all();
187
188		paravirt_release_pte(__pa(pte) >> PAGE_SHIFT);
189		pte = newpte;
190	}
191	BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1)
192	       && vaddr > fix_to_virt(FIX_KMAP_END)
193	       && lastpte && lastpte + PTRS_PER_PTE != pte);
194#endif
195	return pte;
196}
197
198/*
199 * This function initializes a certain range of kernel virtual memory
200 * with new bootmem page tables, everywhere page tables are missing in
201 * the given range.
202 *
203 * NOTE: The pagetables are allocated contiguous on the physical space
204 * so we can cache the place of the first one and move around without
205 * checking the pgd every time.
206 */
207static void __init
208page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
209{
210	int pgd_idx, pmd_idx;
211	unsigned long vaddr;
212	pgd_t *pgd;
213	pmd_t *pmd;
214	pte_t *pte = NULL;
215	unsigned long count = page_table_range_init_count(start, end);
216	void *adr = NULL;
217
218	if (count)
219		adr = alloc_low_pages(count);
220
221	vaddr = start;
222	pgd_idx = pgd_index(vaddr);
223	pmd_idx = pmd_index(vaddr);
224	pgd = pgd_base + pgd_idx;
225
226	for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
227		pmd = one_md_table_init(pgd);
228		pmd = pmd + pmd_index(vaddr);
229		for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
230							pmd++, pmd_idx++) {
231			pte = page_table_kmap_check(one_page_table_init(pmd),
232						    pmd, vaddr, pte, &adr);
233
234			vaddr += PMD_SIZE;
235		}
236		pmd_idx = 0;
237	}
238}
239
240static inline int is_kernel_text(unsigned long addr)
241{
242	if (addr >= (unsigned long)_text && addr <= (unsigned long)__init_end)
243		return 1;
244	return 0;
245}
246
247/*
248 * This maps the physical memory to kernel virtual address space, a total
249 * of max_low_pfn pages, by creating page tables starting from address
250 * PAGE_OFFSET:
251 */
252unsigned long __init
253kernel_physical_mapping_init(unsigned long start,
254			     unsigned long end,
255			     unsigned long page_size_mask)
 
256{
257	int use_pse = page_size_mask == (1<<PG_LEVEL_2M);
258	unsigned long last_map_addr = end;
259	unsigned long start_pfn, end_pfn;
260	pgd_t *pgd_base = swapper_pg_dir;
261	int pgd_idx, pmd_idx, pte_ofs;
262	unsigned long pfn;
263	pgd_t *pgd;
264	pmd_t *pmd;
265	pte_t *pte;
266	unsigned pages_2m, pages_4k;
267	int mapping_iter;
268
269	start_pfn = start >> PAGE_SHIFT;
270	end_pfn = end >> PAGE_SHIFT;
271
272	/*
273	 * First iteration will setup identity mapping using large/small pages
274	 * based on use_pse, with other attributes same as set by
275	 * the early code in head_32.S
276	 *
277	 * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
278	 * as desired for the kernel identity mapping.
279	 *
280	 * This two pass mechanism conforms to the TLB app note which says:
281	 *
282	 *     "Software should not write to a paging-structure entry in a way
283	 *      that would change, for any linear address, both the page size
284	 *      and either the page frame or attributes."
285	 */
286	mapping_iter = 1;
287
288	if (!boot_cpu_has(X86_FEATURE_PSE))
289		use_pse = 0;
290
291repeat:
292	pages_2m = pages_4k = 0;
293	pfn = start_pfn;
294	pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
295	pgd = pgd_base + pgd_idx;
296	for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
297		pmd = one_md_table_init(pgd);
298
299		if (pfn >= end_pfn)
300			continue;
301#ifdef CONFIG_X86_PAE
302		pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
303		pmd += pmd_idx;
304#else
305		pmd_idx = 0;
306#endif
307		for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn;
308		     pmd++, pmd_idx++) {
309			unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
310
311			/*
312			 * Map with big pages if possible, otherwise
313			 * create normal page tables:
314			 */
315			if (use_pse) {
316				unsigned int addr2;
317				pgprot_t prot = PAGE_KERNEL_LARGE;
318				/*
319				 * first pass will use the same initial
320				 * identity mapping attribute + _PAGE_PSE.
321				 */
322				pgprot_t init_prot =
323					__pgprot(PTE_IDENT_ATTR |
324						 _PAGE_PSE);
325
326				pfn &= PMD_MASK >> PAGE_SHIFT;
327				addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
328					PAGE_OFFSET + PAGE_SIZE-1;
329
330				if (is_kernel_text(addr) ||
331				    is_kernel_text(addr2))
332					prot = PAGE_KERNEL_LARGE_EXEC;
333
334				pages_2m++;
335				if (mapping_iter == 1)
336					set_pmd(pmd, pfn_pmd(pfn, init_prot));
337				else
338					set_pmd(pmd, pfn_pmd(pfn, prot));
339
340				pfn += PTRS_PER_PTE;
341				continue;
342			}
343			pte = one_page_table_init(pmd);
344
345			pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
346			pte += pte_ofs;
347			for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn;
348			     pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
349				pgprot_t prot = PAGE_KERNEL;
350				/*
351				 * first pass will use the same initial
352				 * identity mapping attribute.
353				 */
354				pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR);
355
356				if (is_kernel_text(addr))
357					prot = PAGE_KERNEL_EXEC;
358
359				pages_4k++;
360				if (mapping_iter == 1) {
361					set_pte(pte, pfn_pte(pfn, init_prot));
362					last_map_addr = (pfn << PAGE_SHIFT) + PAGE_SIZE;
363				} else
364					set_pte(pte, pfn_pte(pfn, prot));
365			}
366		}
367	}
368	if (mapping_iter == 1) {
369		/*
370		 * update direct mapping page count only in the first
371		 * iteration.
372		 */
373		update_page_count(PG_LEVEL_2M, pages_2m);
374		update_page_count(PG_LEVEL_4K, pages_4k);
375
376		/*
377		 * local global flush tlb, which will flush the previous
378		 * mappings present in both small and large page TLB's.
379		 */
380		__flush_tlb_all();
381
382		/*
383		 * Second iteration will set the actual desired PTE attributes.
384		 */
385		mapping_iter = 2;
386		goto repeat;
387	}
388	return last_map_addr;
389}
390
391pte_t *kmap_pte;
392
393static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
394{
395	pgd_t *pgd = pgd_offset_k(vaddr);
396	p4d_t *p4d = p4d_offset(pgd, vaddr);
397	pud_t *pud = pud_offset(p4d, vaddr);
398	pmd_t *pmd = pmd_offset(pud, vaddr);
399	return pte_offset_kernel(pmd, vaddr);
400}
401
402static void __init kmap_init(void)
403{
404	unsigned long kmap_vstart;
405
406	/*
407	 * Cache the first kmap pte:
408	 */
409	kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
410	kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
411}
412
413#ifdef CONFIG_HIGHMEM
414static void __init permanent_kmaps_init(pgd_t *pgd_base)
415{
416	unsigned long vaddr;
417	pgd_t *pgd;
418	p4d_t *p4d;
419	pud_t *pud;
420	pmd_t *pmd;
421	pte_t *pte;
422
423	vaddr = PKMAP_BASE;
424	page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
425
426	pgd = swapper_pg_dir + pgd_index(vaddr);
427	p4d = p4d_offset(pgd, vaddr);
428	pud = pud_offset(p4d, vaddr);
429	pmd = pmd_offset(pud, vaddr);
430	pte = pte_offset_kernel(pmd, vaddr);
431	pkmap_page_table = pte;
432}
433
434void __init add_highpages_with_active_regions(int nid,
435			 unsigned long start_pfn, unsigned long end_pfn)
436{
437	phys_addr_t start, end;
438	u64 i;
439
440	for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &start, &end, NULL) {
441		unsigned long pfn = clamp_t(unsigned long, PFN_UP(start),
442					    start_pfn, end_pfn);
443		unsigned long e_pfn = clamp_t(unsigned long, PFN_DOWN(end),
444					      start_pfn, end_pfn);
445		for ( ; pfn < e_pfn; pfn++)
446			if (pfn_valid(pfn))
447				free_highmem_page(pfn_to_page(pfn));
448	}
449}
450#else
451static inline void permanent_kmaps_init(pgd_t *pgd_base)
452{
453}
454#endif /* CONFIG_HIGHMEM */
455
456void __init sync_initial_page_table(void)
457{
458	clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY,
459			swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
460			KERNEL_PGD_PTRS);
461
462	/*
463	 * sync back low identity map too.  It is used for example
464	 * in the 32-bit EFI stub.
465	 */
466	clone_pgd_range(initial_page_table,
467			swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
468			min(KERNEL_PGD_PTRS, KERNEL_PGD_BOUNDARY));
469}
470
471void __init native_pagetable_init(void)
472{
473	unsigned long pfn, va;
474	pgd_t *pgd, *base = swapper_pg_dir;
475	p4d_t *p4d;
476	pud_t *pud;
477	pmd_t *pmd;
478	pte_t *pte;
479
480	/*
481	 * Remove any mappings which extend past the end of physical
482	 * memory from the boot time page table.
483	 * In virtual address space, we should have at least two pages
484	 * from VMALLOC_END to pkmap or fixmap according to VMALLOC_END
485	 * definition. And max_low_pfn is set to VMALLOC_END physical
486	 * address. If initial memory mapping is doing right job, we
487	 * should have pte used near max_low_pfn or one pmd is not present.
488	 */
489	for (pfn = max_low_pfn; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
490		va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
491		pgd = base + pgd_index(va);
492		if (!pgd_present(*pgd))
493			break;
494
495		p4d = p4d_offset(pgd, va);
496		pud = pud_offset(p4d, va);
497		pmd = pmd_offset(pud, va);
498		if (!pmd_present(*pmd))
499			break;
500
501		/* should not be large page here */
502		if (pmd_large(*pmd)) {
503			pr_warn("try to clear pte for ram above max_low_pfn: pfn: %lx pmd: %p pmd phys: %lx, but pmd is big page and is not using pte !\n",
504				pfn, pmd, __pa(pmd));
505			BUG_ON(1);
506		}
507
508		pte = pte_offset_kernel(pmd, va);
509		if (!pte_present(*pte))
510			break;
511
512		printk(KERN_DEBUG "clearing pte for ram above max_low_pfn: pfn: %lx pmd: %p pmd phys: %lx pte: %p pte phys: %lx\n",
513				pfn, pmd, __pa(pmd), pte, __pa(pte));
514		pte_clear(NULL, va, pte);
515	}
516	paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
517	paging_init();
518}
519
520/*
521 * Build a proper pagetable for the kernel mappings.  Up until this
522 * point, we've been running on some set of pagetables constructed by
523 * the boot process.
524 *
525 * If we're booting on native hardware, this will be a pagetable
526 * constructed in arch/x86/kernel/head_32.S.  The root of the
527 * pagetable will be swapper_pg_dir.
528 *
529 * If we're booting paravirtualized under a hypervisor, then there are
530 * more options: we may already be running PAE, and the pagetable may
531 * or may not be based in swapper_pg_dir.  In any case,
532 * paravirt_pagetable_init() will set up swapper_pg_dir
533 * appropriately for the rest of the initialization to work.
534 *
535 * In general, pagetable_init() assumes that the pagetable may already
536 * be partially populated, and so it avoids stomping on any existing
537 * mappings.
538 */
539void __init early_ioremap_page_table_range_init(void)
540{
541	pgd_t *pgd_base = swapper_pg_dir;
542	unsigned long vaddr, end;
543
544	/*
545	 * Fixed mappings, only the page table structure has to be
546	 * created - mappings will be set by set_fixmap():
547	 */
548	vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
549	end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
550	page_table_range_init(vaddr, end, pgd_base);
551	early_ioremap_reset();
552}
553
554static void __init pagetable_init(void)
555{
556	pgd_t *pgd_base = swapper_pg_dir;
557
558	permanent_kmaps_init(pgd_base);
559}
560
561#define DEFAULT_PTE_MASK ~(_PAGE_NX | _PAGE_GLOBAL)
562/* Bits supported by the hardware: */
563pteval_t __supported_pte_mask __read_mostly = DEFAULT_PTE_MASK;
564/* Bits allowed in normal kernel mappings: */
565pteval_t __default_kernel_pte_mask __read_mostly = DEFAULT_PTE_MASK;
566EXPORT_SYMBOL_GPL(__supported_pte_mask);
567/* Used in PAGE_KERNEL_* macros which are reasonably used out-of-tree: */
568EXPORT_SYMBOL(__default_kernel_pte_mask);
569
570/* user-defined highmem size */
571static unsigned int highmem_pages = -1;
572
573/*
574 * highmem=size forces highmem to be exactly 'size' bytes.
575 * This works even on boxes that have no highmem otherwise.
576 * This also works to reduce highmem size on bigger boxes.
577 */
578static int __init parse_highmem(char *arg)
579{
580	if (!arg)
581		return -EINVAL;
582
583	highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
584	return 0;
585}
586early_param("highmem", parse_highmem);
587
588#define MSG_HIGHMEM_TOO_BIG \
589	"highmem size (%luMB) is bigger than pages available (%luMB)!\n"
590
591#define MSG_LOWMEM_TOO_SMALL \
592	"highmem size (%luMB) results in <64MB lowmem, ignoring it!\n"
593/*
594 * All of RAM fits into lowmem - but if user wants highmem
595 * artificially via the highmem=x boot parameter then create
596 * it:
597 */
598static void __init lowmem_pfn_init(void)
599{
600	/* max_low_pfn is 0, we already have early_res support */
601	max_low_pfn = max_pfn;
602
603	if (highmem_pages == -1)
604		highmem_pages = 0;
605#ifdef CONFIG_HIGHMEM
606	if (highmem_pages >= max_pfn) {
607		printk(KERN_ERR MSG_HIGHMEM_TOO_BIG,
608			pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
609		highmem_pages = 0;
610	}
611	if (highmem_pages) {
612		if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) {
613			printk(KERN_ERR MSG_LOWMEM_TOO_SMALL,
614				pages_to_mb(highmem_pages));
615			highmem_pages = 0;
616		}
617		max_low_pfn -= highmem_pages;
618	}
619#else
620	if (highmem_pages)
621		printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
622#endif
623}
624
625#define MSG_HIGHMEM_TOO_SMALL \
626	"only %luMB highmem pages available, ignoring highmem size of %luMB!\n"
627
628#define MSG_HIGHMEM_TRIMMED \
629	"Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n"
630/*
631 * We have more RAM than fits into lowmem - we try to put it into
632 * highmem, also taking the highmem=x boot parameter into account:
633 */
634static void __init highmem_pfn_init(void)
635{
636	max_low_pfn = MAXMEM_PFN;
637
638	if (highmem_pages == -1)
639		highmem_pages = max_pfn - MAXMEM_PFN;
640
641	if (highmem_pages + MAXMEM_PFN < max_pfn)
642		max_pfn = MAXMEM_PFN + highmem_pages;
643
644	if (highmem_pages + MAXMEM_PFN > max_pfn) {
645		printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL,
646			pages_to_mb(max_pfn - MAXMEM_PFN),
647			pages_to_mb(highmem_pages));
648		highmem_pages = 0;
649	}
650#ifndef CONFIG_HIGHMEM
651	/* Maximum memory usable is what is directly addressable */
652	printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20);
653	if (max_pfn > MAX_NONPAE_PFN)
654		printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
655	else
656		printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
657	max_pfn = MAXMEM_PFN;
658#else /* !CONFIG_HIGHMEM */
659#ifndef CONFIG_HIGHMEM64G
660	if (max_pfn > MAX_NONPAE_PFN) {
661		max_pfn = MAX_NONPAE_PFN;
662		printk(KERN_WARNING MSG_HIGHMEM_TRIMMED);
663	}
664#endif /* !CONFIG_HIGHMEM64G */
665#endif /* !CONFIG_HIGHMEM */
666}
667
668/*
669 * Determine low and high memory ranges:
670 */
671void __init find_low_pfn_range(void)
672{
673	/* it could update max_pfn */
674
675	if (max_pfn <= MAXMEM_PFN)
676		lowmem_pfn_init();
677	else
678		highmem_pfn_init();
679}
680
681#ifndef CONFIG_NEED_MULTIPLE_NODES
682void __init initmem_init(void)
683{
684#ifdef CONFIG_HIGHMEM
685	highstart_pfn = highend_pfn = max_pfn;
686	if (max_pfn > max_low_pfn)
687		highstart_pfn = max_low_pfn;
688	printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
689		pages_to_mb(highend_pfn - highstart_pfn));
690	high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
691#else
692	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
693#endif
694
695	memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
696	sparse_memory_present_with_active_regions(0);
697
698#ifdef CONFIG_FLATMEM
699	max_mapnr = IS_ENABLED(CONFIG_HIGHMEM) ? highend_pfn : max_low_pfn;
700#endif
701	__vmalloc_start_set = true;
702
703	printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
704			pages_to_mb(max_low_pfn));
705
706	setup_bootmem_allocator();
707}
708#endif /* !CONFIG_NEED_MULTIPLE_NODES */
709
710void __init setup_bootmem_allocator(void)
711{
712	printk(KERN_INFO "  mapped low ram: 0 - %08lx\n",
713		 max_pfn_mapped<<PAGE_SHIFT);
714	printk(KERN_INFO "  low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT);
715}
716
717/*
718 * paging_init() sets up the page tables - note that the first 8MB are
719 * already mapped by head.S.
720 *
721 * This routines also unmaps the page at virtual kernel address 0, so
722 * that we can trap those pesky NULL-reference errors in the kernel.
723 */
724void __init paging_init(void)
725{
726	pagetable_init();
727
728	__flush_tlb_all();
729
730	kmap_init();
731
732	/*
733	 * NOTE: at this point the bootmem allocator is fully available.
734	 */
735	olpc_dt_build_devicetree();
736	sparse_memory_present_with_active_regions(MAX_NUMNODES);
737	sparse_init();
738	zone_sizes_init();
739}
740
741/*
742 * Test if the WP bit works in supervisor mode. It isn't supported on 386's
743 * and also on some strange 486's. All 586+'s are OK. This used to involve
744 * black magic jumps to work around some nasty CPU bugs, but fortunately the
745 * switch to using exceptions got rid of all that.
746 */
747static void __init test_wp_bit(void)
748{
749	char z = 0;
750
751	printk(KERN_INFO "Checking if this processor honours the WP bit even in supervisor mode...");
752
753	__set_fixmap(FIX_WP_TEST, __pa_symbol(empty_zero_page), PAGE_KERNEL_RO);
754
755	if (probe_kernel_write((char *)fix_to_virt(FIX_WP_TEST), &z, 1)) {
756		clear_fixmap(FIX_WP_TEST);
757		printk(KERN_CONT "Ok.\n");
758		return;
759	}
760
761	printk(KERN_CONT "No.\n");
762	panic("Linux doesn't support CPUs with broken WP.");
763}
764
765void __init mem_init(void)
766{
767	pci_iommu_alloc();
768
769#ifdef CONFIG_FLATMEM
770	BUG_ON(!mem_map);
771#endif
772	/*
773	 * With CONFIG_DEBUG_PAGEALLOC initialization of highmem pages has to
774	 * be done before memblock_free_all(). Memblock use free low memory for
775	 * temporary data (see find_range_array()) and for this purpose can use
776	 * pages that was already passed to the buddy allocator, hence marked as
777	 * not accessible in the page tables when compiled with
778	 * CONFIG_DEBUG_PAGEALLOC. Otherwise order of initialization is not
779	 * important here.
780	 */
781	set_highmem_pages_init();
782
783	/* this will put all low memory onto the freelists */
784	memblock_free_all();
785
786	after_bootmem = 1;
787	x86_init.hyper.init_after_bootmem();
788
789	mem_init_print_info(NULL);
790	printk(KERN_INFO "virtual kernel memory layout:\n"
791		"    fixmap  : 0x%08lx - 0x%08lx   (%4ld kB)\n"
792		"  cpu_entry : 0x%08lx - 0x%08lx   (%4ld kB)\n"
793#ifdef CONFIG_HIGHMEM
794		"    pkmap   : 0x%08lx - 0x%08lx   (%4ld kB)\n"
795#endif
796		"    vmalloc : 0x%08lx - 0x%08lx   (%4ld MB)\n"
797		"    lowmem  : 0x%08lx - 0x%08lx   (%4ld MB)\n"
798		"      .init : 0x%08lx - 0x%08lx   (%4ld kB)\n"
799		"      .data : 0x%08lx - 0x%08lx   (%4ld kB)\n"
800		"      .text : 0x%08lx - 0x%08lx   (%4ld kB)\n",
801		FIXADDR_START, FIXADDR_TOP,
802		(FIXADDR_TOP - FIXADDR_START) >> 10,
803
804		CPU_ENTRY_AREA_BASE,
805		CPU_ENTRY_AREA_BASE + CPU_ENTRY_AREA_MAP_SIZE,
806		CPU_ENTRY_AREA_MAP_SIZE >> 10,
807
808#ifdef CONFIG_HIGHMEM
809		PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
810		(LAST_PKMAP*PAGE_SIZE) >> 10,
811#endif
812
813		VMALLOC_START, VMALLOC_END,
814		(VMALLOC_END - VMALLOC_START) >> 20,
815
816		(unsigned long)__va(0), (unsigned long)high_memory,
817		((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
818
819		(unsigned long)&__init_begin, (unsigned long)&__init_end,
820		((unsigned long)&__init_end -
821		 (unsigned long)&__init_begin) >> 10,
822
823		(unsigned long)&_etext, (unsigned long)&_edata,
824		((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
825
826		(unsigned long)&_text, (unsigned long)&_etext,
827		((unsigned long)&_etext - (unsigned long)&_text) >> 10);
828
829	/*
830	 * Check boundaries twice: Some fundamental inconsistencies can
831	 * be detected at build time already.
832	 */
833#define __FIXADDR_TOP (-PAGE_SIZE)
834#ifdef CONFIG_HIGHMEM
835	BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE	> FIXADDR_START);
836	BUILD_BUG_ON(VMALLOC_END			> PKMAP_BASE);
837#endif
838#define high_memory (-128UL << 20)
839	BUILD_BUG_ON(VMALLOC_START			>= VMALLOC_END);
840#undef high_memory
841#undef __FIXADDR_TOP
842
843#ifdef CONFIG_HIGHMEM
844	BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE	> FIXADDR_START);
845	BUG_ON(VMALLOC_END				> PKMAP_BASE);
846#endif
847	BUG_ON(VMALLOC_START				>= VMALLOC_END);
848	BUG_ON((unsigned long)high_memory		> VMALLOC_START);
849
850	test_wp_bit();
851}
852
853#ifdef CONFIG_MEMORY_HOTPLUG
854int arch_add_memory(int nid, u64 start, u64 size,
855			struct mhp_restrictions *restrictions)
856{
857	unsigned long start_pfn = start >> PAGE_SHIFT;
858	unsigned long nr_pages = size >> PAGE_SHIFT;
859
860	return __add_pages(nid, start_pfn, nr_pages, restrictions);
861}
862
863void arch_remove_memory(int nid, u64 start, u64 size,
864			struct vmem_altmap *altmap)
865{
866	unsigned long start_pfn = start >> PAGE_SHIFT;
867	unsigned long nr_pages = size >> PAGE_SHIFT;
868	struct zone *zone;
869
870	zone = page_zone(pfn_to_page(start_pfn));
871	__remove_pages(zone, start_pfn, nr_pages, altmap);
872}
873#endif
874
875int kernel_set_to_readonly __read_mostly;
876
877void set_kernel_text_rw(void)
878{
879	unsigned long start = PFN_ALIGN(_text);
880	unsigned long size = PFN_ALIGN(_etext) - start;
881
882	if (!kernel_set_to_readonly)
883		return;
884
885	pr_debug("Set kernel text: %lx - %lx for read write\n",
886		 start, start+size);
887
888	set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
889}
890
891void set_kernel_text_ro(void)
892{
893	unsigned long start = PFN_ALIGN(_text);
894	unsigned long size = PFN_ALIGN(_etext) - start;
895
896	if (!kernel_set_to_readonly)
897		return;
898
899	pr_debug("Set kernel text: %lx - %lx for read only\n",
900		 start, start+size);
901
902	set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
903}
904
905static void mark_nxdata_nx(void)
906{
907	/*
908	 * When this called, init has already been executed and released,
909	 * so everything past _etext should be NX.
910	 */
911	unsigned long start = PFN_ALIGN(_etext);
912	/*
913	 * This comes from is_kernel_text upper limit. Also HPAGE where used:
914	 */
915	unsigned long size = (((unsigned long)__init_end + HPAGE_SIZE) & HPAGE_MASK) - start;
916
917	if (__supported_pte_mask & _PAGE_NX)
918		printk(KERN_INFO "NX-protecting the kernel data: %luk\n", size >> 10);
919	set_memory_nx(start, size >> PAGE_SHIFT);
920}
921
922void mark_rodata_ro(void)
923{
924	unsigned long start = PFN_ALIGN(_text);
925	unsigned long size = (unsigned long)__end_rodata - start;
926
927	set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
928	pr_info("Write protecting kernel text and read-only data: %luk\n",
929		size >> 10);
930
931	kernel_set_to_readonly = 1;
932
933#ifdef CONFIG_CPA_DEBUG
934	pr_info("Testing CPA: Reverting %lx-%lx\n", start, start + size);
935	set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
936
937	pr_info("Testing CPA: write protecting again\n");
938	set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
939#endif
940	mark_nxdata_nx();
941	if (__supported_pte_mask & _PAGE_NX)
942		debug_checkwx();
943}